FULL PAPER

Four-Dimensional Respiratory Motion-Resolved WholeHeart Coronary MR Angiography

Davide Piccini,1,2*y Li Feng,3y Gabriele Bonanno,2 Simone Coppo,2 J�erome Yerly,2,4

Ruth P. Lim,5 Juerg Schwitter,6 Daniel K. Sodickson,3 Ricardo Otazo,3 and

Matthias Stuber2,4

Purpose: Free-breathing whole-heart coronary MR angiography

(MRA) commonly uses navigators to gate respiratory motion,

resulting in lengthy and unpredictable acquisition times. Con-

versely, self-navigation has 100% scan efficiency, but requires

motion correction over a broad range of respiratory displace-

ments, which may introduce image artifacts. We propose

replacing navigators and self-navigation with a respiratory

motion-resolved reconstruction approach.

Methods: Using a respiratory signal extracted directly from the

imaging data, individual signal-readouts are binned according to

their respiratory states. The resultant series of undersampled

images are reconstructed using an extradimensional golden-angle

radial sparse parallel imaging (XD-GRASP) algorithm, which

exploits sparsity along the respiratory dimension. Whole-heart cor-

onary MRA was performed in 11 volunteers and four patients with

the proposed methodology. Image quality was compared with that

obtained with one-dimensional respiratory self-navigation.Results: Respiratory-resolved reconstruction effectively sup-

pressed respiratory motion artifacts. The quality score for

XD-GRASP reconstructions was greater than or equal to self-

navigation in 80/88 coronary segments, reaching diagnostic qual-

ity in 61/88 segments versus 41/88. Coronary sharpness and

length were always superior for the respiratory-resolved datasets,

reaching statistical significance (P < 0.05) in most cases.

Conclusion: XD-GRASP represents an attractive alternative for

handling respiratory motion in free-breathing whole heart MRI

and provides an effective alternative to self-navigation. MagnReson Med 77:1473–1484, 2017. VC 2016 International Societyfor Magnetic Resonance in Medicine

Key words: coronary MRA; free breathing; sparse reconstruc-tion; compressed sensing; motion correction; self-navigation

INTRODUCTION

Coronary MR angiography (MRA) has improved substan-tially over the past two decades, showing promising resultsin relatively large patient cohorts (1–4). However, it stillremains mostly confined to research use at a small numberof experienced academic MR centers. The main shortcom-ing of coronary MRA is its vulnerability to motion-inducedartifacts, which is linked with time-inefficient data collec-tion. Conventional acquisitions are electrocardiogram-triggered to short time windows (�50–100 ms) of relativecoronary quiescence (e.g., mid-diastole), and a large numberof cardiac cycles is needed for sufficient high-resolutionvolumetric coverage. Consequently, free-breathing acquisi-tions are mandatory, but they necessitate sophisticatedrespiratory motion compensation approaches.

Navigator techniques (5–8), which identify the posi-tion of the right hemidiaphragm, typically gate datacollection to a small acceptance window during end-expiration only. The drawbacks of these techniquesinclude low scan efficiency, unpredictable scan times,sensitivity to respiratory drift, and the need for meticu-lous scan planning. Conversely, self-navigated techni-ques (9–14) that directly extract a respiratory motionsignal from the imaging data to perform respiratorymotion correction, operate with 100% scan efficiency,have highly predictable scan duration, and improvedease-of-use. Such techniques have been extensivelytested in both volunteer (10,15,16) and patient (14,17)cohorts in single-center studies. However, self-navigationstill suffers from limitations related to the motion modelemployed for correction (e.g., one-dimensional [1D],three-dimensional [3D], affine, etc.) (18), may not be suf-ficiently accurate for large-scale respiratory motion, andmay be affected by artifacts originating from static anat-omy after motion correction.

Several 3D self-navigation approaches have been pro-posed recently to subdivide the whole heart coronarydataset into subsets of volumes, each corresponding to aparticular respiratory position or state (12,13,19,20).Because the quality of these imaging volumes is generallylow due to the inevitable undersampling artifacts, allsuch methods propose to exploit these subsets uniquelyfor image registration, while the final reconstructed data-set is composed of all acquired k-space lines after motion

1Advanced Clinical Imaging Technology, Siemens Healthcare, Lausanne,Switzerland.2Department of Radiology, University Hospital and University of Lausanne,Lausanne, Switzerland.3Center for Advanced Imaging Innovation and Research, and Bernard andIrene Schwartz Center for Biomedical Imaging, Department of Radiology,New York University School of Medicine, New York, New York, USA.4Center for Biomedical Imaging, Lausanne, Switzerland.5Department of Radiology, Austin Health and The University of Melbourne,Melbourne, Victoria, Australia.6Division of Cardiology and Cardiac MR Center, University Hospital ofLausanne, Lausanne, Switzerland.

Grant sponsor: Center for Advanced Imaging Innovation and Research(CAI2R, www.cai2r.net), a NIBIB Biomedical Technology Resource Center;Grant number: P41 EB017183; Grant sponsor: Swiss National ScienceFoundation; Grant number: 143923.

*Correspondence to: Davide Piccini Ph.D., Center for BioMedical Imaging,Centre Hospitalier Universitaire Vaudois, Rue de Bugnon 46, BH 7.84, 1011Lausanne, Switzerland. E-mail: piccinidavide@gmail.com; Twitter:@CVMR_Lausanne

yDavide Piccini and Li Feng contributed equally to this study.

Received 29 October 2015; revised 25 January 2016; accepted 24February 2016

DOI 10.1002/mrm.26221Published online 28 March 2016 in Wiley Online Library (wileyonlinelibrary.com).

Magnetic Resonance in Medicine 77:1473–1484 (2017)

VC 2016 International Society for Magnetic Resonance in Medicine 1473

correction. In this context, it has to be mentioned that theregistration of undersampled volumes is a nontrivial prob-lem, due, among other technical issues, to the artifactlevel connected to the undersampling. However, with theintroduction of sparse reconstruction techniques, whichexploit correlations between images in the series toreduce the number of k-space points required to recon-struct each individual image (21,22), the image quality ofsuch subsets may potentially be enhanced such thatimage registration and corrected reconstruction are nolonger be needed.

Recently, a novel image reconstruction framework

known as extradimensional golden-angle radial sparse

parallel (XD-GRASP) MRI was proposed, which com-

bines the acceleration capabilities of reduced k-space

sampling and sparse reconstruction with the self-

navigation properties of a hybrid radial and Cartesian

(stack-of-stars) sampling scheme to reconstruct addi-

tional motion dimensions for dynamic liver imaging and

cine cardiac imaging (23). 3D whole heart coronary MRA

is a static rather than dynamic acquisition, but it can be

an ideal extension for XD-GRASP, because the data

acquired during free-breathing can be sorted and binned

into multiple respiratory phases. This generates a fourth

dimension, which represents the respiratory motion. As

in the original XD-GRASP implementation, the new

respiratory dimension can be regarded as an additional

temporal correlation that can be exploited using a

sparsity-based approach to reconstruct a four-

dimensional (4D) dataset (3D þ respiratory dimension).

Nevertheless, to extend this approach and achieve the

whole-heart coverage and high isotropic spatial resolu-

tion needed for free-breathing coronary imaging, a

recently proposed (24) 3D golden angle radial sampling

scheme was used. This sampling scheme not only ena-

bles the flexible a posteriori combination of k-space seg-

ments needed for the respiratory binning, it also

provides additional incoherence when compared with

the original XD-GRASP implementation.We therefore hypothesize that such sparse reconstruc-

tion algorithms can be exploited to reconstruct respira-

tory motion-resolved 3D images of the heart without the

need for breath-holding, navigators, self-navigated respi-

ratory correction, or complex 3D motion correction.

METHODS

A general schematic representation of the acquisition

and reconstruction pipeline is shown in Figure 1. Details

about each step of the pipeline are provided in the sec-

tions below.

Data Acquisition

Written informed consent was obtained from all partici-

pants for this Institutional Review Board–approved

study. Volunteer and patient data acquisitions were per-

formed using a prototype segmented 3D radial trajectory(24). Such a trajectory is subjected to a golden angle rota-

tion about the z-axis at each heartbeat, such that each

acquired segment is intrinsically positioned within the

largest k-space void left by the prior segments. Addition-

ally, a readout with consistent superior–inferior (SI) ori-

entation at the beginning of each segment can be used

for either respiratory self-navigation (10) or respiratory

data binning (25). Because of the unique golden angle

rotational arrangement of the readouts, it is possible to

retrospectively sort (i.e., bin) all segments according to

the respiratory phases (motion states) at which they were

acquired. This results in distinct pseudo-uniform 3D k-

space coverage for each respiratory bin.Free-breathing whole heart coronary MRA was per-

formed in 11 healthy adult volunteers (male, n ¼ 9;

female, n ¼ 2; mean age, 29 6 4 y [range, 26–34 y]) and,

as a first proof of principle, in four patients (male, n ¼ 3;

female, n ¼ 1; mean age, 64 6 7 y [range, 54–72 y]; sinus

rhythm range 55–92 beats per minute) with established

coronary artery disease. The acquisition was performed

using electrocardiogram triggering on a 1.5T clinical MRI

scanner (MAGNETOM Aera; Siemens Healthcare,

Erlangen, Germany) with a total of 30 receiver coil ele-

ments. For the volunteer group, the acquisition window(�80–100 ms, �25–32 k-space lines per cardiac cycle)

was placed in mid-diastole and adapted to the individ-

ual subject’s heart rate using visual inspection of a four-

chamber cine dataset at high temporal resolution while

the scan was performed during approximately 400 heart-

beats. For the patient group, the acquisition window was

adapted to a shorter cardiac rest phase (�40–60 ms/12–20

k-space lines per cardiac cycle), which was compensated

FIG. 1. Schematic representation of the proposed acquisition/reconstruction pipeline. The flow chart shows how the radial k-space linesof the whole heart coronary MRA datasets are acquired with a 3D radial acquisition scheme, as opposed to the 2D radial stack-of-starsscheme proposed in the original publication. The acquired static data are first used for respiratory motion extraction and then conse-

quently binned into distinct respiratory phases via the data sorting procedure. The sorted data are then fed into the XD-GRASP sparseiterative reconstruction algorithm to generate the final respiratory motion-resolved dataset. Coil sensitivity maps are also directly

extracted from the radial acquisition and subsequently used in the iterative reconstruction.

1474 Piccini et al.

for by an increased number of heartbeats (�600). The pro-

tocol parameters of the 3D radial, non–slice-selective, T2-

prepared (echo time T2 prep ¼ 40 ms), fat-saturated, bal-

anced steady state free precession imaging sequence

included: repetition time ¼ 3.1 ms; echo time ¼ 1.56 ms;

field of view ¼ 220 mm3, matrix ¼ 192 � 192 � 192; voxel

size ¼ 1.15 mm3; RF excitation angle ¼ 90�; and receiver

bandwidth ¼ 898 Hz/Pixel. The total amount of k-space

lines acquired in each case was approximately 12,000 for a

total acquisition time of 7.0 6 1.3 min.

Respiratory Motion Extraction

A recently described technique (25) that exploits signal

amplitude variations from the center of k-space (KCA)

(26) was adopted to extract respiratory-induced motion

from all SI-oriented readouts. Because such variations

have different contributions in each detector coil, the

respiratory motion detection procedure (Fig. 2) was

repeated for all coil elements individually. Subsequently,

independent component analysis (27) was performed on

the KCA signals from all coils to extract independent sig-

nal components. The signal component with maximum

amplitude, within the bandwidth of the expected respi-

ratory frequency (28), was used for respiratory binning.

Data Sorting

Figure 3a illustrates the data sorting procedure prior to XD-

GRASP reconstruction. The whole heart coronary MRA data,

acquired using the 3D radial trajectory, were subdivided into

n subsampled 3D k-space datasets that originate from n dif-

ferent respiratory states. The number of k-space lines

grouped in each individual state was the same. According to

the golden angle acquisition scheme, nearly uniform cover-

age of k-space is ensured for each bin and distinct sampling

patterns among those bins are automatically obtained so that

temporal incoherence and sparsity can be exploited in the

iterative reconstruction. For the current study, four and six

respiratory phases were used for data binning (Fig. 3b,c).

Although a six-phase subdivision of the breathing pattern

leads to narrower and more sparsely populated bins (12,13),

a four-phase subdivision leads to wider bins that are more

densely populated. Both four- and six-phase XD-GRASP

reconstructions were performed to investigate the effects of

bin width on image quality.According to prior experience (23), the regularization

parameters of the XD-GRASP reconstruction were empiri-

cally optimized for four- and six-phase XD-GRASP on two

representative training datasets by two coronary MRA

experts (D.P. and M.S., with 7 and 20 years of experience

in coronary MRA, respectively) blinded to the reconstruc-

tion parameters. Specifically, a range of values (0.005–0.1)

of the regularization parameter were tested and compared

visually. The values that generated the best image quality

and coronary delineation were selected separately for the

four- and six-phase XD-GRASP reconstructions.Because the binning process is driven by the amount

of data contained in each bin, the bin widths are not

constant. Consequently, the amount of motion included

in each bin is also variable within the same subject. The

relative amount of motion included in each bin or respi-

ratory phase for the two binning approaches (four- and

six-phase) was measured as the percentage width of each

bin relative to the entire respiratory excursion of a sub-

ject, and these bin widths (expressed in %) were then

averaged across all subjects. All values were reported as

the mean 6 standard deviation.

Image Reconstruction

XD-GRASP reconstruction was performed on all binned

coronary datasets by solving the following optimization

problem that enforces joint multicoil sparsity on the

extradimensional dataset using a sparsifying transform

along the new respiratory dimension

d ¼ argmind

jjF � C � d �mjj22 þ ljjS � djj1; [1]

where F is the nonuniform fast Fourier transform

(NUFFT) operator defined for the 3D golden angle radial

sampling pattern; C represents the multiple-element

coil sensitivity maps with dimensionality of the x–y–z–

coil (where x, y, and z represent the three spatial

dimensions); d is the respiratory motion-resolved image

FIG. 2. Procedure for respiratory motion detection. First, separate components of the KCA signal are extracted from the SI readouts for

each receiver coil element (left). Subsequently, independent component analysis is performed on all KCA signals to identify differentindependent components (center). The component with highest amplitude in the frequency range of respiration (right, black line) is thenselected and used as the respiratory motion signal for data sorting into respiratory bins.

Respiratory Motion-Resolved Coronary MRA 1475

series with dimensions x–y–z–t; and m represents the cor-

responding multicoil radial k-space data, which are sorted

according to the respiratory state with dimension x–y–z–t–

coil. The sparsifying transform S is the first-order differ-

ence between respiratory states and the minimization of its

l1 norm is commonly known as total variation minimiza-

tion. This approach is well suited for applications where

the image or volume is piecewise constant, and it has per-

formed well when compared with other recently proposed

sparsifying transforms (29). Finally, k is the regularization

parameter that weighs the contribution of the joint multi-

coil sparse regularization with respect to data consistency.

Image reconstruction was implemented as described in

Feng et al. (23), and the most end-expiratory volume was

chosen for image analysis.For comparison, each coronary dataset was also recon-

structed using 1D respiratory self-navigation as described

by Piccini et al. (14). For respiratory self-navigation, respi-

ratory motion was extracted by cross-correlating the auto-

matically segmented blood pool on the 1D FFT of an end-

expiratory reference SI readout (reference SI projection)

(15) to all other SI projections. The resulting 1D SI dis-

placement was then used for correcting each data segment

before the gridding operation by directly applying a phase

shift to all k-space readouts and was adapted for the polar

orientation of each signal readout. Image reconstruction

was performed using a 3D regridding algorithm that makes

use of a uniform density compensation function and a Kai-

ser–Bessel window. The multicoil 3D images were then

combined using coil sensitivity maps estimated using the

adaptive combination approach described in Walsh et al.

(30). A detailed description of the segmentation and cross-

correlation algorithms can be found in Piccini et al. (14).All reconstructions were performed using a server

equipped with two 16-core Opteron CPUs, 256 Gb RAM,

and two 6-Gb NVIDIA graphics processing unit (GPU)

cards. The NUFFT operation was implemented using

parallel computing on GPUs (31), and the main recon-

struction program was implemented in MATLAB 2012b

(MathWorks, Natick, Massachusetts, USA).

FIG. 3. Data sorting procedure for XD-GRASP reconstruction. (a) The whole heart coronary MRA dataset acquired using the phyllotaxis 3Dradial trajectory can be binned into any number (n) of respiratory phases, spanning from end-expiration (top) to end-inspiration (bottom),

using the motion signal detected. Each dot represents the origin (on the top half of k-space) of a linear readout going through the centerof k-space. The binning procedure is then performed so that the number of spokes grouped in each respiratory phase is equal, therefore

leading to different bin widths. (b, c) Due to the golden angle acquisition scheme, approximately uniform coverage of k-space with distinctsampling patterns in each phase is always achieved. Two examples of data binning with four (b) and six (c) separate phases are shown.

1476 Piccini et al.

Volunteer Study

All 1D respiratory self-navigated and end-expiratory phasesof XD-GRASP reconstructed volunteer datasets were pooled,anonymized, and randomized prior to image analysis.

Quantitative Assessment

To objectively determine the quality of vessel depiction,the proximal and mid-segments of both the right coronaryartery (RCA) and the left anterior descending coronaryartery (LAD) as well as the left main stem (LM) and theproximal left circumflex coronary artery (LCX) were identi-fied on all datasets, as described by Austen et al. (32). Thepercentage vessel sharpness was computed for each ofthese segments as described by Etienne et al. (33). Thetotal visible vessel lengths were also measured for both theleft coronary system (sum of LM and LAD) and the RCA.

Image Quality Assessment

On all datasets, the image quality of the entire LM; theproximal, mid-, and distal segments of the LAD andRCA; and the proximal LCX were graded on a 5-pointscale similar to that described by McConnell et al (34).The two aforementioned experts performed this grading(image quality grading) using consensus reading (3). Thescale classified the definition of the vessel borders as fol-lows: 0 ¼ not visible, 1 ¼ markedly blurred, 2 ¼ moder-ately blurred, 3 ¼ mildly blurred, and 4 ¼ sharplydefined. The average values of the quality grading wererecorded, as were the number of equal or higher scoresin either of the two techniques.

Diagnostic Quality Assessment

The diagnostic quality and visibility of each segment ofLM, LAD, RCA, and proximal LCX were graded for allcoronary datasets on a 3-point scale (diagnostic qualitygrading) by an experienced cardiovascular MR radiologist(R.P.L., with 10 years of experience), blinded to the typeof reconstruction. The scale used was as follows: 0 ¼ notvisible, 1 ¼ visible but nondiagnostic, and 2 ¼ diagnostic.

Statistical Analysis

Statistical comparisons for the vessel sharpness andlength were performed with a paired two-tailed Studentt test in Microsoft Excel (Microsoft, Redmond, Washing-ton, USA) and P < 0.05 was considered statistically sig-nificant. For quality grading and diagnostic grading, anonparametric paired two-tailed Wilcoxon signed-ranktest was used to compare the scores on a per-vessel seg-ment basis, as well as all scores for all segments of allpatients considered together. Finally, a McNemar’s statis-tical test was used to assess statistical significance forthe number of coronary segments that were graded asdiagnostic with the three reconstructions. Bonferroni cor-rection was applied where needed to correct for the mul-tiple comparisons.

Initial Data Collection in Patients

Patient data were reconstructed using the four-phase XD-GRASP reconstruction, and because of the small sample

size, the results were only visually compared with 1D

respiratory self-navigation. Identification of the locationof coronary stenoses on MRI was also visually compared

with the findings from the gold standard X-ray coronary

angiography.

RESULTS

Data Sorting

A four-phase XD-GRASP reconstruction was found to besufficient for resolving respiratory motion, and a regula-

rization parameter of k ¼ 0.02 led to an optimal balance

between removal of undersampling artifacts and recon-structed image quality. A six-phase XD-GRASP recon-

struction achieved similar performance with a higher

regularization parameter (k ¼ 0.05) to account for the

higher degree of undersampling. The relative amount ofrespiratory displacement for each of the bins was the

smallest for the end-expiratory phase, with 14.6% 6

4.8% and 8.2% 6 2.8% of the total motion for the four-phase and six-phase reconstruction, respectively. The

phases with largest respiratory displacement were meas-

ured around end-inspiration at 35.3% 6 8.0% and27.1% 6 7.4% for four-phase and six-phase XD-GRASP,

respectively. Figure 4 shows images from one representa-

tive volunteer in whom conventional gridding recon-

structions with sensitivity-weighted multicoilcombination for each respiratory phase, after binning,

were visually compared with those obtained from four-

phase XD-GRASP. As evidenced by a gradual shift ofboth the heart and the liver in these images, respiratory

motion can be resolved and accounted for by sorting the

data into different motion states, whereas undersamplingartifacts are efficiently removed with XD-GRASP recon-

struction. Image quality of the motion states improved

considerably using XD-GRASP, whereas the image regis-

tration step was no longer needed.

Image Reconstruction

The whole heart coronary MRA acquisitions and recon-

structions were successful for all volunteer and patientscans. The average reconstruction time for XD-GRASP

was 35.1 6 4.5 min and 58.2 6 5.6 min for four-phase

and six-phase XD-GRASP, respectively. By comparison,

the 1D respiratory self-navigation inline reconstructionon the scanner lasted <2 min.

Volunteer Study

An overview of the numerical results of image quality

grading, vessel sharpness, and length measurements isreported in Table 1.

Quantitative Assessment

Greater or equal vessel sharpness with respect to 1D respi-

ratory self-navigation was obtained in 56/66 and 49/66 cor-onary segments, and the total visible vessel length was

superior or equal in 17/22 and 15/22 evaluated coronary

arteries for the four- and six-phase XD-GRASP reconstruc-tions, respectively. The average vessel sharpness and

length were superior for both XD-GRASP reconstructions

Respiratory Motion-Resolved Coronary MRA 1477

with respect to 1D respiratory self-navigation. Four-phase

XD-GRASP vessel sharpness was significantly superior to

1D respiratory self-navigation for the LM (P < 0.003), prox-

imal, and mid-LAD (P < 0.04 and P < 0.007) and mid-

RCA (P < 0.008). Visualized length of LMþLAD (P < 0.05)

was also improved. In addition, significantly superior

results for the six-phase XD-GRASP reconstruction were

achieved for the sharpness values of the mid-RCA (P <

0.005) compared with 1D respiratory self-navigation. Figure

5 shows multiplanar reformats of the coronary arteries

from one representative volunteer: although 1D respiratory

self-navigation already led to adequate image quality (Fig.

5a), there was a clear improvement in sharpness and visi-

ble vessel length in both the four-phase and six-phase XD-

GRASP reconstructions.

Image Quality Assessment

In the vast majority of cases, the respiratory-resolved XD-GRASP reconstruction achieved equal or improved imagequality grades relative to 1D respiratory self-navigation. Inparticular, the quality grade was greater or equal in 76/88and in 80/88 coronary segments for the four- and six-phaseXD-GRASP reconstructions, respectively. The average gradeof the qualitative assessment was superior for both XD-GRASP reconstructions, except for the quality grade of theproximal LCX, which was very similar to that of 1D respira-tory self-navigation. The paired two-tailed Wilcoxon signed-rank test showed statistical significance for the improve-ments in the image quality grades of all coronary segmentsconsidered together for both four-phase XD-GRASP recon-struction (P < 0.0001) and six-phase XD-GRASP

FIG. 4. Example from one representative healthy adult volunteer comparing conventional sensitivity-weighted multicoil gridding recon-struction with the four-phase XD-GRASP reconstruction. All reconstructed phases are shown in an axial plane (top) and a coronal plane(bottom). Respiratory motion is resolved by sorting the data into different motion states, while the streaking artifacts due to undersam-

pling are effectively removed in the XD-GRASP reconstruction.

1478 Piccini et al.

reconstruction (P < 0.0001) compared with 1D respiratoryself-navigation. There was no statistically significant differ-ence (P¼ 0.06) between the two XD-GRASP reconstructions.When looking at the per segment analysis, statistical signifi-cance was reached for the mid-LAD for the four-phase XD-GRASP and for the mid-LAD, mid-RCA, and distal RCA forsix-phase XD-GRASP (Table 1).

Diagnostic Quality Assessment

The results related to the diagnostic grading are reportedin Table 2. Overall, the ratio of coronary segments labeledas diagnostic increased from 41/88 for 1D respiratory self-navigation to 61/88 and 56/88 for the four- and six-phaseXD-GRASP reconstructions, respectively. Conversely, thetotal ratio of nonvisible segments decreased from 5/88with 1D respiratory-self navigation to 1/88 with both four-and six-phase XD-GRASP reconstructions.

The diagnostic grades obtained for the four-phase XD-GRASP reconstructions were higher or equal to those of1D respiratory self-navigation in 87/88 coronary segments,whereas this ratio was 85/88 for six-phase XD-GRASP. Onaverage, the diagnostic grades for the XD-GRASP recon-structions were always higher than those obtained with1D respiratory self-navigation, with the only exceptionbeing the proximal LCX. The four-phase XD-GRASPreconstruction reached 100% diagnostic quality for LM,proximal LAD, and proximal RCA. Full diagnostic qualitywas also reached with the six-phase reconstruction forLM and proximal RCA.

The paired two-tailed Wilcoxon signed-rank test showedstatistical significance for the improvements in the diagnosticquality grades of all coronary segments considered togetherfor both four-phase XD-GRASP reconstruction (P < 0.0001)and six-phase XD-GRASP reconstruction (P < 0.0005) com-

pared with 1D respiratory self-navigation. The McNemar’sstatistical test showed that the increase in the number of diag-nostic coronary segments was significant for both four-phase(P < 0.0001) and six-phase (P < 0.003) reconstructions withrespect to 1D respiratory self-navigation. Also in this case, acomparison between the four-phase and six-phase approachdid not yield statistically significant results.

A volunteer dataset in which 1D respiratory self-navigation did not achieve diagnostic quality is shown inFigure 6. Although there was a visual improvement withrespect to the uncorrected reconstruction (where gridding ofall k-space lines was performed without any respiratorymotion correction), the coronal view showed substantialresidual blurring, and only the proximal segments of LADand LCX were visible. Even though a long contiguous seg-ment of the RCA was displayed, diagnostic quality was notreached for the more distal segments. By contrast, using XD-GRASP reconstruction, sharp and well-defined borders ofcoronary, myocardial, and hepatic structures can be appreci-ated. Coronary segments that were graded as visible but non-diagnostic on the 1D respiratory self-navigation wereconsidered diagnostic after XD-GRASP reconstruction.

Initial Findings in Patients

All datasets were successfully reconstructed in this ini-tial and small patient cohort. All patients had at leasttwo-vessel coronary artery disease in the proximal ormid-segments as diagnosed by X-ray coronary angiogra-phy. In three of the four patients, one or more coronarystenoses were successfully visualized by way of MR. Theimage quality of the fourth patient was not sufficient toconfirm stenoses in the mid- and distal coronary seg-ments. Importantly, the regularization term of the four-phase XD-GRASP reconstruction did not affect the

Table 1Quality Grading and Quantitative Assessment

Coronary Segment Measured Parameter 1D Self-Navigation

XD-GRASP

Four-Phase Six-Phase

LM Image quality grading 1.8 6 0.8 2.0 6 0.5 2.1 6 0.5Sharpness (%) 48.4 6 7.7 55.1 6 7.7a 49.1 6 8.5

LAD (proximal) Image quality grading 1.5 6 0.8 1.9 6 0.5 2.0 6 0.6Sharpness (%) 37.3 6 8.6 41.9 6 8.1a 40.2 6 7.7

LAD (mid) Image quality grading 1.0 6 0.8 1.5 6 0.4a 1.5 6 0.6a

Sharpness (%) 34.9 6 10.2 40.8 6 9.1a 38.7 6 8.7LAD (distal) Image quality grading 0.8 6 0.6 1.1 6 0.6 1.2 6 0.8

LMþLAD Length (cm) 11.7 6 3.2 13.2 6 3.1a 12.8 6 3.1a

LCX (proximal) Image quality grading 1.1 6 0.7 1.0 6 0.5 1.0 6 0.5

Sharpness (%) 43.2 6 9.3 46.3 6 12.9 48.7 6 7.9RCA (proximal) Image quality grading 1.9 6 0.7 2.2 6 0.4 2.3 6 0.5

Sharpness (%) 46.9 6 11.0 53.2 6 5.3 52.5 6 5.6

RCA (mid) Image quality grading 1.7 6 0.6 2.0 6 0.4 2.2 6 0.7a

Sharpness (%) 42.9 6 10.4 49.8 6 6.2a 49.9 6 7.5a

RCA (distal) Image quality grading 1.2 6 0.8 1.5 6 0.4 1.7 6 0.7a

RCA Length (cm) 11.2 6 2.5 12.1 6 1.6 11.7 6 3.1

All values are expressed as the mean 6 standard deviation of the image quality grades, vessel sharpness, and length for the 11 volun-

teers and using 1D respiratory self-navigation compared with the proposed respiratory-resolved XD-GRASP reconstruction. The imagequality grading system was as follows: 0 ¼ not visible, 1 ¼ markedly blurred, 2 ¼ moderately blurred, 3 ¼ mildly blurred, and 4 ¼sharply defined.Abbreviations: LAD, left anterior descending artery; LCX, left circumflex artery; LM, left main stem; RCA, right coronary artery; XD-GRASP, extradimensional golden-angle radial sparse parallel MRI.aIndicates statistical significance compared with 1D respiratory self-navigation.

Respiratory Motion-Resolved Coronary MRA 1479

depiction of stenosis, and the vessels appeared sharper

than in the reformats from 1D respiratory self-navigation.

Example reformats from one coronary artery disease

patient are shown in Figure 7; both 1D respiratory self-

navigation and four-phase XD-GRASP reconstruction are

shown in comparison to the gold standard X-ray coro-

nary angiogram. A significant stenosis of the proximal

LAD was well visualized in both the self-navigated and

the XD-GRASP reconstruction.

DISCUSSION

XD-GRASP has been introduced previously as a sparse

iterative reconstruction technique that can be exploited

for motion suppression in dynamic MRI using a stack-of-

stars sampling scheme (23). In the present study, the XD-

GRASP framework was extended for free-breathing, 3D

whole heart coronary MRA. For this purpose, it was

combined with a 3D golden-angle radial trajectory that

enables not only whole heart coverage at high isotropic

spatial resolution but also a flexible a posteriori combi-

nation of k-space segments that are regrouped into

pseudo-uniform phases or bins.Specifically, we tested the hypothesis that this

approach can be leveraged to reconstruct respiratory

motion-resolved 3D images of the heart without the need

for navigators, self-navigated respiratory correction, or

tailored models for 3D motion correction. A respiratory

signal is extracted directly from the imaging data and is

used only for sorting the acquired data into multiple

respiratory states. Although conventional gridding

reconstruction of the individual bins may provide

motion-resolved images, the considerable degree of under-

sampling would inevitably lead to noise and streaking

artifacts. However, the correlation along the new respira-

tory dimension is well suited for reconstruction using the

FIG. 5. Examples of multiplanar reformatted coronary arteries from one representative healthy adult volunteer. Although respiratory self-navigated reconstruction with 1D motion correction could achieve high image quality (top row), a clear improvement in sharpness as

well as visible vessel length (arrows) can be noticed in both four-phase (middle row) and six-phase (bottom row) XD-GRASPreconstructions.

1480 Piccini et al.

XD-GRASP framework. With this approach, image regis-

tration of low-quality sub-images can be avoided, as the

respiratory dimension is used to create a sparse time

domain that can be exploited to enhance the image qual-

ity of each individual respiratory bin. Although the end-

expiratory 3D volume of the XD-GRASP reconstruction

was selected for coronary artery visualization in our

study, it should be noted that the reconstructed motion-

resolved datasets contain abundant anatomic information

across the whole respiratory cycle, which may be

exploited further (Supporting Video 1).This approach was tested in a group of healthy adult

volunteers and preliminarily in a small number of

patients with coronary artery disease and was compared

with 1D respiratory self-navigation. In this context, it is

important to emphasize that the exact same acquired

data were used, yet the reconstruction pathway was dif-

ferent. The results show that vessel conspicuity and ves-

sel length were superior with the proposed methodology.

The image quality assessment was used to identify the

overall improvements in image quality and was per-

formed by the first and last authors of the manuscript,

who have the most experience in coronary MRA.

Although this might be a potential source of bias, the

results were fully confirmed by the diagnostic quality

assessment, which was performed by an independent

reviewer and answered the question of whether diagnos-

tic quality was obtained with each technique. Further-

more, initial patient data also demonstrated considerable

promise for improved delineation of proximal coronary

segments.Because conventional whole heart coronary MRA is

commonly performed with end-expiratory navigator gat-

ing that suffers from variable scan efficiency and unpre-

dictable scanning times, several groups have proposed

the acquisition of data during the entire respiratorycycle, together with aggressive motion correction toobtain reduced and predictable scan times (9–11). Inmost cases, respiratory motion is retrospectively cor-rected using registration algorithms (12,13,18,35). How-ever, the motion correction performed using specificmotion models (11,13,36,37) remains an approximationof the real respiratory-induced motion of the heart.Hypothetically, this limitation promises to be intrinsi-cally addressed using the XD-GRASP methodology,because neither motion models nor registration algo-rithms are required. Nevertheless, a direct comparisonbetween the proposed motion-resolved reconstructionapproach and established techniques based on navigatorgating will be required to test this hypothesis. However,the 1D self-navigation technique has already been com-pared with the standard navigator-gated technique involunteers (10), and similar image quality was obtained.For this reason, and because of scan time constraints, adirect comparison between the XD-GRASP techniqueand a standard navigator-gated technique was not addedto the present study.

An important aspect also relates to the tradeoffbetween the number of reconstructed respiratory phasesand the residual, intra-bin, sparsity artifacts. Here, XD-GRASP reconstruction was performed with both fourand six respiratory phases. Although the use of a largernumber of respiratory phases might be expected to betterresolve respiratory motion and reduce blurring, a highernumber of phases is also associated with increaseddegrees of undersampling in each of them. The resultssuggest that, for our healthy volunteer cohort, both four-and six-phase XD-GRASP reconstructions are very simi-lar in performance. Ideally, the number of reconstructedrespiratory phases could be tailored to the specificbreathing pattern of each subject—choosing, for example,less phases for more regular breathers, where the overallrespiratory displacement is relatively small, and morephases for irregular breathers, where the total respiratoryamplitude may be larger (38). Although the resultsobtained in the initial patient cohort seem to confirmthat a four-phase reconstruction might be sufficient alsofor this population, the number of subjects is too smalland the link between respiratory drift or irregular breath-ing and an optimal reconstruction yet remains to beascertained. Further analyses on a larger patient popula-tion would be useful to investigate the relationshipbetween the different breathing patterns and the optimalnumber of reconstructed phases.

Although sparse reconstruction techniques have beenapplied previously for whole heart coronary MRA(20,39–43), most of the implementations aim at reducingthe acquisition time by exploiting spatial correlations.XD-GRASP, on the other hand, sorts the acquired datainto an additional respiratory dimension and exploitssparsity in this new dynamic dimension. The high corre-lation in this respiratory dimension enables improvedperformance in XD-GRASP reconstruction comparedwith conventional sparse reconstructions exploiting spa-tial correlation only.

The golden-angle arrangement of the 3D radial phyllo-taxis trajectory intrinsically facilitates the extraction of

Table 2Diagnostic Quality Grading

CoronarySegment

1D RespiratorySelf-Navigation

XD-GRASP

Four-Phase Six-Phase

LM 1.8 6 0.4 2.0 6 0.0 2.0 6 0.0LAD (proximal) 1.6 6 0.5 2.0 6 0.0 1.7 6 0.5

LAD (mid) 1.3 6 0.6 1.4 6 0.5 1.5 6 0.5LAD (distal) 0.9 6 0.5 1.3 6 0.5 1.3 6 0.5

LCX (proximal) 1.4 6 0.7 1.4 6 0.7 1.4 6 0.7RCA (proximal) 1.8 6 0.4 2.0 6 0.0 2.0 6 0.0RCA (mid) 1.3 6 0.5 1.7 6 0.5 1.5 6 0.5

RCA (distal) 1.4 6 0.7 1.7 6 0.5 1.6 6 0.5Total diagnostic

segments

41/88 (47%) 61/88 (70%)a 56/88 (63%)a

All values are expressed as mean 6 standard deviation for diag-nostic quality grades for the 11 volunteers, using 1D respiratory

self-navigation compared with the proposed XD-GRASP recon-struction. A clear improvement was seen in the datasets where

respiratory-resolved XD-GRASP reconstruction was applied. Thediagnostic quality grading system was as follows: 0 ¼ not visible,1 ¼ visible but nondiagnostic, and 2 ¼ visible and diagnostic.

Abbreviations: LAD, left anterior descending artery; LCX, left cir-cumflex artery; LM, left main stem; RCA, right coronary artery; XD-

GRASP, extradimensional golden-angle radial sparse parallel MRI.aIndicates statistical significance compared with 1D respiratoryself-navigation.

Respiratory Motion-Resolved Coronary MRA 1481

the respiratory motion signal directly from the imagedata (since all readouts cross the k-space center) and pro-vides flexibility in data sorting (since quasi-uniform spa-tial sampling over time—and therefore in all respiratorybins—is facilitated). Simultaneously, the golden anglerotation ensures a certain level of incoherence along thenew respiratory dimension, which is required for XD-GRASP reconstruction (i.e., the data are sufficiently dis-tinct in different motion states, since the same k-spaceprofile is never acquired twice).

In electrocardiogram-triggered whole heart coronaryMRA, data acquisition is typically performed during ashort acquisition window in diastole. However, 4D car-diac and respiratory motion-resolved whole heart coro-nary MRA approaches have been proposed recently(44,45) that advocate continuous data acquisitionthrough the entire cardiac cycle. Coronary artery imagescan be reconstructed at different cardiac phases, andadditional information on 3D cardiac function can beextracted. The extension of these 4D approaches withXD-GRASP seems straightforward, and it would enable

simultaneous exploitation of correlations in both cardiac

and respiratory dimensions, thus leading to 5D cardiac

MRI.Two limitations of our initial proof-of-concept imple-

mentation include: 1) lengthy reconstruction times due

to the iterative reconstruction process, which requires

the computation of one forward and one inverse 3D

NUFFT for each respiratory phase and coil per iteration,

and 2) substantial memory requirements to respond to

the computational needs associated with large, high-

resolution 3D datasets. However, with continued pro-

gress in computer hardware and software, this is not

likely to be a future roadblock. Although this study had

a strong focus on coronary MRA with most stringent

boundary conditions because of the small and tortuous

nature of the vessels, the XD-GRASP approach will

likely benefit other 3D cardiac imaging techniques such

as delayed hyperenhancement studies and T1 and T2

mapping. Finally, XD-GRASP may potentially be com-

bined with methods that simultaneously exploit the

FIG. 6. Example of a selected healthy adult volunteer dataset in which 1D respiratory self-navigation did not lead to diagnostic image

quality. Although a clear improvement can be noticed from the uncorrected reconstruction to 1D respiratory self-navigation, the coronalview still shows strong residual blurring after conventional respiratory motion correction. Only the proximal segments of LAD and LCX

are visible, whereas the RCA, although visible, was not scored with full diagnostic quality in the mid- and distal segments. By contrast,the dataset provided by the XD-GRASP reconstruction shows sharp and well-defined margins of a papillary muscle in the left ventricleand of the liver on the coronal view (arrowheads). All coronary arteries were better depicted, and increased vessel length could be seen

with XD-GRASP reconstruction. Coronary segments graded as visible but nondiagnostic in the 1D respiratory self-navigation were con-sidered diagnostic with the proposed XD-GRASP methodology (arrows).

1482 Piccini et al.

spatial sparsity of the images. Based on earlier numeri-

cal simulations (29), only a moderate improvement in

image quality is expected and an increasing degree of

freedom for optimization of the regularization weights

will have to be considered.

CONCLUSION

A powerful alternative to navigators or self-navigation

for handling respiratory motion in 3D whole heart coro-

nary MRA has been proposed. The XD-GRASP frame-

work can be exploited to reconstruct respiratory motion-

resolved 3D images of the heart without the need for

breath-holding, navigators, self-navigated respiratory

motion correction, or complex 3D correction schemes.Instead of discarding data segments from inconsistent

respiratory phases or enforcing motion models for motion

correction that are inevitably imperfect, XD-GRASP makes

constructive use of all respiratory phases to improve

image quality and achieves superior image quality com-

pared with 1D respiratory self-navigation. A 3D radial

phyllotaxis trajectory and XD-GRASP reconstruction pro-

vide a synergistic combination that ultimately may lead

coronary MRA closer to clinical practice.

ACKNOWLEDGMENTS

We would like to thank Dr. Florian Knoll from NYU

School of Medicine for his support with the GPU imple-

mentation of 3D NUFFT.

REFERENCES

1. Kato S, Kitagawa K, Ishida N, et al. Assessment of coronary artery

disease using magnetic resonance coronary angiography: a national

multicenter trial. J Am Coll Cardiol 2010;56:983–991.

2. Sakuma H. Coronary CT versus MR angiography: the role of MR angi-

ography. Radiology 2011;258:340–349.

3. Kim WY, Danias PG, Stuber M, et al. Coronary magnetic resonance

angiography for the detection of coronary stenoses. N Engl J Med

2001;345:1863–1869.

4. Hamdan A, Asbach P, Wellnhofer E, Klein C, Gebker R, Kelle S,

Kilian H, Huppertz A, Fleck E. A prospective study for comparison

of MR and CT imaging for detection of coronary artery stenosis. JACC

Cardiovasc Imaging 2011;4:50–61.

5. Ehman RL, McNamara MT, Pallack M, Hricak H, Higgins CB. Mag-

netic resonance imaging with respiratory gating: techniques and

advantages. Am J Roentgenol 1984;143:1175–1182.

6. Wang Y, Rossman PJ, Grimm RC, Riederer SJ, Ehman RL. Navigator-

echo-based real-time respiratory gating and triggering for reduction of

respiration effects in three-dimensional coronary MR angiography.

Radiology 1996;198:55–60.

7. Danias PG, McConnell MV, Khasgiwala VC, Chuang ML, Edelman

RR, Manning WJ. Prospective navigator correction of image position

for coronary MR angiography. Radiology 1997;203:733–736.

8. Stuber M, Botnar RM, Danias PG, Sodickson DK, Kissinger KV, Van

Cauteren M, De Becker J, Manning WJ. Double-oblique free-breathing

high resolution three-dimensional coronary magnetic resonance angi-

ography. J Am Coll Cardiol 1999;34:524–531.

9. Stehning C, Bornert P, Nehrke K, Eggers H, Stuber M. Free-breathing

whole-heart coronary MRA with 3D radial SSFP and self-navigated

image reconstruction. Magn Reson Med 2005;54:476–480.

10. Piccini D, Littmann A, Nielles-Vallespin S, Zenge MO. Respiratory

self-navigation for whole-heart bright-blood coronary MRI: methods

for robust isolation and automatic segmentation of the blood pool.

Magn Reson Med 2012;68:571–579.

11. Henningsson M, Koken P, Stehning C, Razavi R, Prieto C, Botnar RM.

Whole-heart coronary MR angiography with 2D self-navigated image

reconstruction. Magn Reson Med 2012;67:437–445.

12. Bhat H, Ge L, Nielles-Vallespin S, Zuehlsdorff S, Li D. 3D radial sam-

pling and 3D affine transform-based respiratory motion correction

technique for free-breathing whole-heart coronary MRA with 100%

imaging efficiency. Magn Reson Med 2011;65:1269–1277.

13. Pang J, Bhat H, Sharif B, Fan Z, Thomson LE, Labounty T, Friedman JD,

Min J, Berman DS, Li D. Whole-heart coronary MRA with 100% respira-

tory gating efficiency: self-navigated three-dimensional retrospective

image-based motion correction (TRIM). Magn Reson Med 2014;71:67–74.

14. Piccini D, Monney P, Sierro C, et al. Respiratory self-navigated post-

contrast whole-heart coronary MR angiography: initial experience in

patients. Radiology 2014;270:378–386.

15. Piccini D, Bonanno G, Ginami G, Littmann A, Zenge MO, Stuber M.

Is there an optimal respiratory reference position for self-navigated

whole-heart coronary MR angiography? J Magn Reson Imaging 2016;

43:426–433.

16. Renker M, Varga-Szemes A, Schoepf UJ, et al. A non-contrast self-

navigated 3-dimensional MR technique for aortic root and vascular

access route assessment in the context of transcatheter aortic valve

replacement: proof of concept. Eur Radiol 2016;26:951–958.

FIG. 7. Sample images from a selected patient dataset. Both 1D respiratory self-navigation and the four-phase XD-GRASP reconstruc-tion can be compared directly with the standard 2D X-ray angiogram. The X-ray angiogram shows a significant stenosis of the proximal

LAD, which is also well visualized in the corresponding self-navigated dataset (arrows). The four-phase XD-GRASP reconstruction doesnot affect depiction of the stenosis (arrow), whereas the vessel itself appears to be even sharper. The signal void present on MR images

distal to the stenosis is caused by the presence of a coronary stent in the proximal LCX.

Respiratory Motion-Resolved Coronary MRA 1483

17. Monney P, Piccini D, Rutz T, et al. Single centre experience of the

application of self navigated 3D whole heart cardiovascular magnetic

resonance for the assessment of cardiac anatomy in congenital heart

disease. J Cardiovasc Magn Reson 2015;17:55.

18. Manke D, Nehrke K, Bornert P, Rosch P, Dossel O. Respiratory

motion in coronary magnetic resonance angiography: a comparison of

different motion models. J Magn Reson Imaging 2002;15:661–671.

19. Prieto C, Doneva M, Usman M, Henningsson M, Greil G, Schaeffter

T, Botnar RM. Highly efficient respiratory motion compensated free-

breathing coronary mra using golden-step Cartesian acquisition.

J Magn Reson Imaging 2015;41:738–746.

20. Bonanno G, Puy G, Wiaux Y, van Heeswijk RB, Piccini D, Stuber M.

Self-navigation with compressed sensing for 2D translational motion

correction in free-breathing coronary MRI: a feasibility study. PloS

One 2014;9:e105523.

21. Lustig M, Donoho D, Pauly JM. Sparse MRI: the application of com-

pressed sensing for rapid MR imaging. Magn Reson Med 2007;58:

1182–1195.

22. Otazo R, Kim D, Axel L, Sodickson DK. Combination of compressed

sensing and parallel imaging for highly accelerated first-pass cardiac

perfusion MRI. Magn Reson Med 2010;64:767–776.

23. Feng L, Axel L, Chandarana H, Block KT, Sodickson DK, Otazo R.

XD-GRASP: golden-angle radial MRI with reconstruction of extra

motion-state dimensions using compressed sensing. Magn Reson Med

2016;75:775–788.

24. Piccini D, Littmann A, Nielles-Vallespin S, Zenge MO. Spiral phyllo-

taxis: the natural way to construct a 3D radial trajectory in MRI.

Magn Reson Med 2011;66:1049–1056.

25. Bonanno G, Piccini D, Mar�echal B, Zenge MO, Stuber M. A New Bin-

ning Approach for 3D Motion Corrected Self-Navigated Whole-Heart

Coronary MRA Using Independent Component Analysis of Individual

Coils. In Proceedings of the 22nd Annual Meeting of ISMRM, Milan,

Italy, 2014. p. 936.

26. Larson AC, White RD, Laub G, McVeigh ER, Li D, Simonetti OP. Self-

gated cardiac cine MRI. Magn Reson Med 2004;51:93–102.

27. Hyvarinen A. Fast and robust fixed-point algorithms for independent

component analysis. IEEE Trans Neural Netw 1999;10:626–634.

28. Buehrer M, Curcic J, Boesiger P, Kozerke S. Prospective self-gating

for simultaneous compensation of cardiac and respiratory motion.

Magn Reson Med 2008;60:683–690.

29. Feng L, Sirchai MB, Lim RP, Harrison A, King W, Adluru G, Dibella

EVR, Sodickson DK, Otazo R, Kim D. Highly accelerated real-time

cardiac cine MRI using k-t SPARSE-SENSE. Magn Reson Med 2013;

70:64–74.

30. Walsh DO, Gmitro AF, Marcellin MW. Adaptive reconstruction of

phased array MR imagery. Magn Reson Med 2000;43:682–690.

31. Knoll F, Schwarzl A, Diwoky C, Sodickson DK. gpuNUFFT - An

Open Source GPU Library for 3D Regridding with Direct MATLAB

Interface. In Proceedings of the 22nd Annual Meeting of ISMRM,

Milan, Italy, 2014. p. 4297.

32. Austen WG, Edwards JE, Frye RL, Gensini GG, Gott VL, Griffith LS,

McGoon DC, Murphy ML, Roe BB. A reporting system on patients

evaluated for coronary artery disease. Report of the Ad Hoc Commit-

tee for Grading of Coronary Artery Disease, Council on Cardiovascu-

lar Surgery, American Heart Association. Circulation 1975;51:5–40.

33. Etienne A, Botnar RM, Van Muiswinkel AM, Boesiger P, Manning

WJ, Stuber M. “Soap-Bubble” visualization and quantitative analysis

of 3D coronary magnetic resonance angiograms. Magn Reson Med

2002;48:658–666.

34. McConnell MV, Khasgiwala VC, Savord BJ, Chen MH, Chuang ML,

Edelman RR, Manning WJ. Comparison of respiratory suppression

methods and navigator locations for MR coronary angiography. Am J

Roentgenol 1997;168:1369–1375.

35. Schmidt JF, Buehrer M, Boesiger P, Kozerke S. Nonrigid retrospective

respiratory motion correction in whole-heart coronary MRA. Magn

Reson Med 2011;66:1541–1549.

36. Aitken AP, Henningsson M, Botnar RM, Schaeffter T, Prieto C. 100%

Efficient three-dimensional coronary MR angiography with two-

dimensional beat-to-beat translational and bin-to-bin affine motion

correction. Magn Reson Med 2015;74:756–764.

37. Ingle RR, Wu HH, Addy NO, Cheng JY, Yang PC, Hu BS,

Nishimura DG. Nonrigid autofocus motion correction for coronary

MR angiography with a 3D cones trajectory. Magn Reson Med

2014;72:347–361.

38. Ginami G, Bonanno G, Schwitter J, Stuber M, Piccini D. An iterative

approach to respiratory self-navigated whole-heart coronary MRA sig-

nificantly improves image quality in a preliminary patient study.

Magn Reson Med 2016;75:1594–1604.

39. Akcakaya M, Basha TA, Chan RH, Manning WJ, Nezafat R. Acceler-

ated isotropic sub-millimeter whole-heart coronary MRI: compressed

sensing versus parallel imaging. Magn Reson Med 2014;71:815–822.

40. Akcakaya M, Nam S, Hu P, Moghari MH, Ngo LH, Tarokh V,

Manning WJ, Nezafat R. Compressed sensing with wavelet domain

dependencies for coronary MRI: a retrospective study. IEEE Trans

Med Imaging 2011;30:1090–1099.

41. Moghari MH, Akcakaya M, O’Connor A, et al. Compressed-sensing

motion compensation (CosMo): a joint prospective-retrospective

respiratory navigator for coronary MRI. Magn Reson Med 2011;66:

1674–1681.

42. Forman C, Piccini D, Grimm R, Hutter J, Hornegger J, Zenge MO.

High-resolution 3D whole-heart coronary MRA: a study on the combi-

nation of data acquisition in multiple breath-holds and 1D residual

respiratory motion compensation. Magn Reson Mater Phy 2014;27:

435–443.

43. Forman C, Piccini D, Grimm R, Hutter J, Hornegger J, Zenge MO.

Reduction of respiratory motion artifacts for free-breathing whole-

heart coronary MRA by weighted iterative reconstruction. Magn

Reson Med 2014;73:1885–1895.

44. Pang J, Sharif B, Fan Z, Bi X, Arsanjani R, Berman DS, Li D. ECG and

navigator-free four-dimensional whole-heart coronary MRA for simul-

taneous visualization of cardiac anatomy and function. Magn Reson

Med 2014;72:1208–1217.

45. Coppo S, Piccini D, Bonanno G, Chaptinel J, Vincenti G, Feliciano H,

van Heeswijk RB, Schwitter J, Stuber M. Free-running 4D whole-

heart self-navigated golden angle MRI: initial results. Magn Reson

Med 2015;74:1306–1316.

SUPPORTING INFORMATION

Additional Supporting Information may be found in the online version ofthis article.

Video S1. Example of a complete respiratory-resolved whole heart datasetfrom one of the volunteers. The animation scrolls through all slices of thedataset in the three standard orientations: (a) transversal, (b) coronal, and(c) sagittal, while all respiratory motion states are displayed for each slice.Although only the most end-expiratory state was used for the qualitativeand quantitative comparisons, the XD-GRASP reconstruction provides onevolume with high and isotropic spatial resolution for each respiratorymotion state.

1484 Piccini et al.